A Systematic Study of Cross-Layer KV Sharing for Efficient LLM Inference

Recently, sharing key-value (KV) cache across layers has been found effective in efficient inference of large language models (LLMs). To systematically investigate different techniques of cross-layer KV sharing, we propose a unified framework that covers several recent methods and their novel variants. We conduct comprehensive experiments on all the configurations of the framework, evaluating their generation throughput and performance in language modeling and downstream tasks. We find that when reducing the size of the KV cache by 2x, most configurations can achieve competitive performance to and higher throughput than standard transformers, but when further reducing the size of the KV cache, pairing queries of all layers with KVs of upper layers can better maintain performance, although it also introduces additional training cost and prefilling latency. We hope that this work will help users choose the appropriate approach according to their requirements and facilitate research on the acceleration of LLM inference.

Efficient Long-range Language Modeling with Self-supervised Causal Retrieval

Recently, retrieval-based language models (RLMs) have received much attention. However, most of them leverage a pre-trained retriever with fixed parameters, which may not adapt well to causal language models. In this work, we propose Grouped Cross-Attention, a novel module enabling joint pre-training of the retriever and causal LM, and apply it to long-context modeling. For a given input sequence, we split it into chunks and use the current chunk to retrieve past chunks for subsequent text generation. Our innovation allows the retriever to learn how to retrieve past chunks that better minimize the auto-regressive loss of subsequent tokens in an end-to-end manner. By integrating top-$k$ retrieval, our model can be pre-trained efficiently from scratch with context lengths up to 64K tokens. Our experiments show our model, compared with long-range LM baselines, can achieve lower perplexity with comparable or lower pre-training and inference costs.

Learning Robust Named Entity Recognizers From Noisy Data With Retrieval Augmentation

Named entity recognition (NER) models often struggle with noisy inputs, such as those with spelling mistakes or errors generated by Optical Character Recognition processes, and learning a robust NER model is challenging. Existing robust NER models utilize both noisy text and its corresponding gold text for training, which is infeasible in many real-world applications in which gold text is not available. In this paper, we consider a more realistic setting in which only noisy text and its NER labels are available. We propose to retrieve relevant text of the noisy text from a knowledge corpus and use it to enhance the representation of the original noisy input. We design three retrieval methods: sparse retrieval based on lexicon similarity, dense retrieval based on semantic similarity, and self-retrieval based on task-specific text. After retrieving relevant text, we concatenate the retrieved text with the original noisy text and encode them with a transformer network, utilizing self-attention to enhance the contextual token representations of the noisy text using the retrieved text. We further employ a multi-view training framework that improves robust NER without retrieving text during inference. Experiments show that our retrieval-augmented model achieves significant improvements in various noisy NER settings.

Dependency Transformer Grammars: Integrating Dependency Structures into Transformer Language Models

Syntactic Transformer language models aim to achieve better generalization through simultaneously modeling syntax trees and sentences. While prior work has been focusing on adding constituency-based structures to Transformers, we introduce Dependency Transformer Grammars (DTGs), a new class of Transformer language model with explicit dependency-based inductive bias. DTGs simulate dependency transition systems with constrained attention patterns by modifying attention masks, incorporate the stack information through relative positional encoding, and augment dependency arc representation with a combination of token embeddings and operation embeddings. When trained on a dataset of sentences annotated with dependency trees, DTGs achieve better generalization while maintaining comparable perplexity with Transformer language model baselines. DTGs also outperform recent constituency-based models, showing that dependency can better guide Transformer language models. Our code is released at https://github.com/zhaoyd1/Dep_Transformer_Grammars.

Sparser is Faster and Less is More: Efficient Sparse Attention for Long-Range Transformers

Accommodating long sequences efficiently in autoregressive Transformers, especially within an extended context window, poses significant challenges due to the quadratic computational complexity and substantial KV memory requirements inherent in self-attention mechanisms. In this work, we introduce SPARSEK Attention, a novel sparse attention mechanism designed to overcome these computational and memory obstacles while maintaining performance. Our approach integrates a scoring network and a differentiable top-k mask operator, SPARSEK, to select a constant number of KV pairs for each query, thereby enabling gradient-based optimization. As a result, SPARSEK Attention offers linear time complexity and constant memory footprint during generation. Experimental results reveal that SPARSEK Attention outperforms previous sparse attention methods and provides significant speed improvements during both training and inference, particularly in language modeling and downstream tasks. Furthermore, our method can be seamlessly integrated into pre-trained Large Language Models (LLMs) with minimal fine-tuning, offering a practical solution for effectively managing long-range dependencies in diverse applications.

Unsupervised Morphological Tree Tokenizer

As a cornerstone in language modeling, tokenization involves segmenting text inputs into pre-defined atomic units. Conventional statistical tokenizers often disrupt constituent boundaries within words, thereby corrupting semantic information. To address this drawback, we introduce morphological structure guidance to tokenization and propose a deep model to induce character-level structures of words. Specifically, the deep model jointly encodes internal structures and representations of words with a mechanism named $\textit{MorphOverriding}$ to ensure the indecomposability of morphemes. By training the model with self-supervised objectives, our method is capable of inducing character-level structures that align with morphological rules without annotated training data. Based on the induced structures, our algorithm tokenizes words through vocabulary matching in a top-down manner. Empirical results indicate that the proposed method effectively retains complete morphemes and outperforms widely adopted methods such as BPE and WordPiece on both morphological segmentation tasks and language modeling tasks. The code will be released later.

Layer-Condensed KV Cache for Efficient Inference of Large Language Models

Huge memory consumption has been a major bottleneck for deploying high-throughput large language models in real-world applications. In addition to the large number of parameters, the key-value (KV) cache for the attention mechanism in the transformer architecture consumes a significant amount of memory, especially when the number of layers is large for deep language models. In this paper, we propose a novel method that only computes and caches the KVs of a small number of layers, thus significantly saving memory consumption and improving inference throughput. Our experiments on large language models show that our method achieves up to 26$\times$ higher throughput than standard transformers and competitive performance in language modeling and downstream tasks. In addition, our method is orthogonal to existing transformer memory-saving techniques, so it is straightforward to integrate them with our model, achieving further improvement in inference efficiency. Our code is available at https://github.com/whyNLP/LCKV.

Potential and Limitations of LLMs in Capturing Structured Semantics: A Case Study on SRL

Large Language Models (LLMs) play a crucial role in capturing structured semantics to enhance language understanding, improve interpretability, and reduce bias. Nevertheless, an ongoing controversy exists over the extent to which LLMs can grasp structured semantics. To assess this, we propose using Semantic Role Labeling (SRL) as a fundamental task to explore LLMs' ability to extract structured semantics. In our assessment, we employ the prompting approach, which leads to the creation of our few-shot SRL parser, called PromptSRL. PromptSRL enables LLMs to map natural languages to explicit semantic structures, which provides an interpretable window into the properties of LLMs. We find interesting potential: LLMs can indeed capture semantic structures, and scaling-up doesn't always mirror potential. Additionally, limitations of LLMs are observed in C-arguments, etc. Lastly, we are surprised to discover that significant overlap in the errors is made by both LLMs and untrained humans, accounting for almost 30% of all errors.

RoT: Enhancing Large Language Models with Reflection on Search Trees

Large language models (LLMs) have demonstrated impressive capability in reasoning and planning when integrated with tree-search-based prompting methods. However, since these methods ignore the previous search experiences, they often make the same mistakes in the search process. To address this issue, we introduce Reflection on search Trees (RoT), an LLM reflection framework designed to improve the performance of tree-search-based prompting methods. It uses a strong LLM to summarize guidelines from previous tree search experiences to enhance the ability of a weak LLM. The guidelines are instructions about solving this task through tree search which can prevent the weak LLMs from making similar mistakes in the past search process. In addition, we proposed a novel state selection method, which identifies the critical information from historical search processes to help RoT generate more specific and meaningful guidelines. In our extensive experiments, we find that RoT significantly improves the performance of LLMs in reasoning or planning tasks with various tree-search-based prompting methods (e.g., BFS and MCTS). Non-tree-search-based prompting methods such as Chain-of-Thought (CoT) can also benefit from RoT guidelines since RoT can provide task-specific knowledge collected from the search experience.

Improving Retrieval Augmented Open-Domain Question-Answering with Vectorized Contexts

In the era of large language models, applying techniques such as Retrieval Augmented Generation can better address Open-Domain Question-Answering problems. Due to constraints including model sizes and computing resources, the length of context is often limited, and it becomes challenging to empower the model to cover overlong contexts while answering questions from open domains. This paper proposes a general and convenient method to covering longer contexts in Open-Domain Question-Answering tasks. It leverages a small encoder language model that effectively encodes contexts, and the encoding applies cross-attention with origin inputs. With our method, the origin language models can cover several times longer contexts while keeping the computing requirements close to the baseline. Our experiments demonstrate that after fine-tuning, there is improved performance across two held-in datasets, four held-out datasets, and also in two In Context Learning settings.